transformers

Finetuning various transformer architectures on a variety of standard NLP datasets using hugging face, pytorch, and other associated frameworks.

Sentiment Analysis

finetuning_sentiment.ipynb:

• Dataset: glue sst2
• Model: distillbert (uncased)

finetuning_sentiment_custom.ipynb:

• Dataset: twitter airline sentiment
• Model: distillbert (cased)

Textual Entailment

finetuning_entailment.ipynb:

• Dataset: glue rte
• Model: bert (cased)

Named Entity Recognition

finetuning_ner.ipynb:

• Dataset: conll2003
• Model: bert (cased)

Parts-of-Speech Tagging

finetuning_pos_custom.ipynb:

• Dataset: brown corpus
• Model: bert (cased)

Neural Machine Translation

finetuning_translation.ipynb:

• Dataset: Helsinki-NLP kde4
• Model: Helsinki-NLP opus-mt-en-hi

---

P.S. This was done on a singular RTX2060Ti GPU, hence the selection of relatively lightweight models and datasets.